Claude Code productivity tricks that nobody talks about

Everyone knows the basics: write code, run tests, fix bugs. But Claude Code has a layer of hidden productivity patterns that change how you work.

Here are the ones I use every day.

1. Ask Claude to explain before it codes

By default, Claude jumps straight into implementation. This burns context on code you might not want.

Instead:

Before you write any code, explain your approach in 3 bullet points. 
I'll tell you if it's right before you start.

This catches misunderstandings in 50 tokens instead of 2000.

2. Use --output-format stream-json for automation

If you're running Claude Code in a script:

claude --output-format stream-json --print "refactor auth.js" \
  | jq -r 'select(.type=="text") | .text'

This gives you machine-readable output you can pipe into your CI pipeline.

3. Pin your tech stack in CLAUDE.md

Stop telling Claude your stack every session. Put it in CLAUDE.md once:

## Tech Stack
- Node.js 22, Express 5
- PostgreSQL with Prisma ORM
- Vitest for testing (NOT Jest)
- Tailwind CSS (NOT Bootstrap)

## Conventions
- Use async/await (NOT .then chains)
- Always add JSDoc to exported functions
- Tests go in __tests__/ folder

Now every session starts with full context. No repetition.

4. The "rubber duck" pattern

Before asking Claude to fix a bug, describe it:

I have a bug where X happens when Y. My current theory is Z. 
Do you see any flaws in my theory before we look at code?

Often Claude spots the issue in your description, saving a full file read.

5. Chain commands with semicolons

claude "add error handling to fetchUser" ; \
claude "write tests for fetchUser" ; \
claude "update the JSDoc for fetchUser"

Each runs sequentially. Useful for known multi-step tasks.

6. Use --allowedTools to restrict scope

claude --allowedTools "Read,Write" "update the README"

This prevents Claude from running shell commands or making git commits when you just want a docs update. Safer for documentation-only tasks.

7. Create project-specific aliases

In your .bashrc or .zshrc:

alias cc-test='claude "run the test suite and fix any failures"'
alias cc-review='claude "review the last 3 commits and flag any issues"'
alias cc-docs='claude "update README based on recent changes"'

Now cc-test is a single command that runs your full TDD loop.

8. Rate limits kill flow — proxy them out

If you hit rate limits mid-session (common on intensive days), your context disappears.

Fix:

export ANTHROPIC_BASE_URL=https://simplylouie.com

This routes Claude Code through SimplyLouie — a $2/month proxy with no rate limits. Your workflow never stops.

# Verify it's working
curl https://simplylouie.com/health
# → {"status":"ok"}

9. Ask for "confidence levels"

When Claude proposes a fix:

How confident are you in this fix on a scale of 1-10? 
What would make you more confident?

Low confidence = look at the code yourself. High confidence = ship it.

10. End every session with a handoff note

Write a 5-bullet summary of what we changed today. 
Include: what we fixed, what we deferred, and what the next logical step is.
Save it to SESSION-NOTES.md.

Next session, you have instant context restoration.

---

The pattern across all of these: reduce friction at every step. Less repetition, less interruption, more flow.

Which of these do you already use? What am I missing? Drop it in the comments.

---

Rate limits interrupting your Claude Code flow? SimplyLouie is a $2/month Claude proxy — set ANTHROPIC_BASE_URL and never hit a limit mid-session again. 7-day free trial.